Kimi K3.1 Pro Goes Live on Inference.net at Launch

Inference.net, which runs AI models behind a pay-per-token API, announced on November 4 that Moonshot AI's Kimi K3.1 Pro was live on its platform from "Day 0" (announcement). Outside hosts can run the model because Moonshot publishes its weights. The model card describes an open-weight model post-trained from Kimi K3, with 2.8 trillion parameters and a context window of about one million tokens, the amount of text it can take in at once. The license allows commercial use, but a business that sells model access through an API and has more than $20 million in total revenue over any 12 months must first sign a separate agreement with Moonshot, unless it is one of Moonshot's certified inference partners.

On Inference.net, the model sits behind an OpenAI-compatible API, so code written for OpenAI's client library can call it by changing the server address and model name. The model page lists image input, tool calling and a setting for how much reasoning the model does before answering; Moonshot's model card lists three levels, low, high and max, with max as the default. Prompt caching needs no extra setup: when calls repeat an identical opening, the repeated part is billed at a lower cache-read rate.

The announcement's first post, the only part of the thread that was collected, claims K3.1 Pro is more capable, cheaper and faster than Kimi K3. Only the price claim can be checked against Inference.net's own page. On November 5 it listed K3.1 Pro at $2 per million input tokens and $10 per million output tokens, against $3 and $15 for Kimi K3, a third lower on both. Inference.net publishes no speed or quality measurements there. Moonshot's card reports its own benchmark results, which this explanation did not evaluate. Nathan Lambert's analysis covers the July Kimi K3 release this model is built from.